2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
31 static struct kmem_cache
*fuse_inode_cachep
;
32 struct list_head fuse_conn_list
;
33 DEFINE_MUTEX(fuse_mutex
);
35 static int set_global_limit(const char *val
, const struct kernel_param
*kp
);
37 unsigned max_user_bgreq
;
38 module_param_call(max_user_bgreq
, set_global_limit
, param_get_uint
,
39 &max_user_bgreq
, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq
, "uint");
41 MODULE_PARM_DESC(max_user_bgreq
,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
45 unsigned max_user_congthresh
;
46 module_param_call(max_user_congthresh
, set_global_limit
, param_get_uint
,
47 &max_user_congthresh
, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh
, "uint");
49 MODULE_PARM_DESC(max_user_congthresh
,
50 "Global limit for the maximum congestion threshold an "
51 "unprivileged user can set");
53 #define FUSE_SUPER_MAGIC 0x65735546
55 #define FUSE_DEFAULT_BLKSIZE 512
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
64 static struct file_system_type fuseblk_fs_type
;
67 struct fuse_forget_link
*fuse_alloc_forget(void)
69 return kzalloc(sizeof(struct fuse_forget_link
), GFP_KERNEL_ACCOUNT
);
72 static struct inode
*fuse_alloc_inode(struct super_block
*sb
)
74 struct fuse_inode
*fi
;
76 fi
= kmem_cache_alloc(fuse_inode_cachep
, GFP_KERNEL
);
87 mutex_init(&fi
->mutex
);
88 spin_lock_init(&fi
->lock
);
89 fi
->forget
= fuse_alloc_forget();
91 kmem_cache_free(fuse_inode_cachep
, fi
);
98 static void fuse_free_inode(struct inode
*inode
)
100 struct fuse_inode
*fi
= get_fuse_inode(inode
);
102 mutex_destroy(&fi
->mutex
);
104 kmem_cache_free(fuse_inode_cachep
, fi
);
107 static void fuse_evict_inode(struct inode
*inode
)
109 struct fuse_inode
*fi
= get_fuse_inode(inode
);
111 truncate_inode_pages_final(&inode
->i_data
);
113 if (inode
->i_sb
->s_flags
& SB_ACTIVE
) {
114 struct fuse_conn
*fc
= get_fuse_conn(inode
);
115 fuse_queue_forget(fc
, fi
->forget
, fi
->nodeid
, fi
->nlookup
);
118 if (S_ISREG(inode
->i_mode
) && !fuse_is_bad(inode
)) {
119 WARN_ON(!list_empty(&fi
->write_files
));
120 WARN_ON(!list_empty(&fi
->queued_writes
));
124 static int fuse_reconfigure(struct fs_context
*fc
)
126 struct super_block
*sb
= fc
->root
->d_sb
;
129 if (fc
->sb_flags
& SB_MANDLOCK
)
136 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
137 * so that it will fit.
139 static ino_t
fuse_squash_ino(u64 ino64
)
141 ino_t ino
= (ino_t
) ino64
;
142 if (sizeof(ino_t
) < sizeof(u64
))
143 ino
^= ino64
>> (sizeof(u64
) - sizeof(ino_t
)) * 8;
147 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
150 struct fuse_conn
*fc
= get_fuse_conn(inode
);
151 struct fuse_inode
*fi
= get_fuse_inode(inode
);
153 lockdep_assert_held(&fi
->lock
);
155 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
156 fi
->i_time
= attr_valid
;
157 WRITE_ONCE(fi
->inval_mask
, 0);
159 inode
->i_ino
= fuse_squash_ino(attr
->ino
);
160 inode
->i_mode
= (inode
->i_mode
& S_IFMT
) | (attr
->mode
& 07777);
161 set_nlink(inode
, attr
->nlink
);
162 inode
->i_uid
= make_kuid(fc
->user_ns
, attr
->uid
);
163 inode
->i_gid
= make_kgid(fc
->user_ns
, attr
->gid
);
164 inode
->i_blocks
= attr
->blocks
;
165 inode
->i_atime
.tv_sec
= attr
->atime
;
166 inode
->i_atime
.tv_nsec
= attr
->atimensec
;
167 /* mtime from server may be stale due to local buffered write */
168 if (!fc
->writeback_cache
|| !S_ISREG(inode
->i_mode
)) {
169 inode
->i_mtime
.tv_sec
= attr
->mtime
;
170 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
171 inode
->i_ctime
.tv_sec
= attr
->ctime
;
172 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
175 if (attr
->blksize
!= 0)
176 inode
->i_blkbits
= ilog2(attr
->blksize
);
178 inode
->i_blkbits
= inode
->i_sb
->s_blocksize_bits
;
181 * Don't set the sticky bit in i_mode, unless we want the VFS
182 * to check permissions. This prevents failures due to the
183 * check in may_delete().
185 fi
->orig_i_mode
= inode
->i_mode
;
186 if (!fc
->default_permissions
)
187 inode
->i_mode
&= ~S_ISVTX
;
189 fi
->orig_ino
= attr
->ino
;
192 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
193 u64 attr_valid
, u64 attr_version
)
195 struct fuse_conn
*fc
= get_fuse_conn(inode
);
196 struct fuse_inode
*fi
= get_fuse_inode(inode
);
197 bool is_wb
= fc
->writeback_cache
;
199 struct timespec64 old_mtime
;
201 spin_lock(&fi
->lock
);
202 if ((attr_version
!= 0 && fi
->attr_version
> attr_version
) ||
203 test_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
)) {
204 spin_unlock(&fi
->lock
);
208 old_mtime
= inode
->i_mtime
;
209 fuse_change_attributes_common(inode
, attr
, attr_valid
);
211 oldsize
= inode
->i_size
;
213 * In case of writeback_cache enabled, the cached writes beyond EOF
214 * extend local i_size without keeping userspace server in sync. So,
215 * attr->size coming from server can be stale. We cannot trust it.
217 if (!is_wb
|| !S_ISREG(inode
->i_mode
))
218 i_size_write(inode
, attr
->size
);
219 spin_unlock(&fi
->lock
);
221 if (!is_wb
&& S_ISREG(inode
->i_mode
)) {
224 if (oldsize
!= attr
->size
) {
225 truncate_pagecache(inode
, attr
->size
);
226 if (!fc
->explicit_inval_data
)
228 } else if (fc
->auto_inval_data
) {
229 struct timespec64 new_mtime
= {
230 .tv_sec
= attr
->mtime
,
231 .tv_nsec
= attr
->mtimensec
,
235 * Auto inval mode also checks and invalidates if mtime
238 if (!timespec64_equal(&old_mtime
, &new_mtime
))
243 invalidate_inode_pages2(inode
->i_mapping
);
247 static void fuse_init_inode(struct inode
*inode
, struct fuse_attr
*attr
)
249 inode
->i_mode
= attr
->mode
& S_IFMT
;
250 inode
->i_size
= attr
->size
;
251 inode
->i_mtime
.tv_sec
= attr
->mtime
;
252 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
253 inode
->i_ctime
.tv_sec
= attr
->ctime
;
254 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
255 if (S_ISREG(inode
->i_mode
)) {
256 fuse_init_common(inode
);
257 fuse_init_file_inode(inode
);
258 } else if (S_ISDIR(inode
->i_mode
))
259 fuse_init_dir(inode
);
260 else if (S_ISLNK(inode
->i_mode
))
261 fuse_init_symlink(inode
);
262 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
263 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
264 fuse_init_common(inode
);
265 init_special_inode(inode
, inode
->i_mode
,
266 new_decode_dev(attr
->rdev
));
271 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
)
273 u64 nodeid
= *(u64
*) _nodeidp
;
274 if (get_node_id(inode
) == nodeid
)
280 static int fuse_inode_set(struct inode
*inode
, void *_nodeidp
)
282 u64 nodeid
= *(u64
*) _nodeidp
;
283 get_fuse_inode(inode
)->nodeid
= nodeid
;
287 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
288 int generation
, struct fuse_attr
*attr
,
289 u64 attr_valid
, u64 attr_version
)
292 struct fuse_inode
*fi
;
293 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
296 inode
= iget5_locked(sb
, nodeid
, fuse_inode_eq
, fuse_inode_set
, &nodeid
);
300 if ((inode
->i_state
& I_NEW
)) {
301 inode
->i_flags
|= S_NOATIME
;
302 if (!fc
->writeback_cache
|| !S_ISREG(attr
->mode
))
303 inode
->i_flags
|= S_NOCMTIME
;
304 inode
->i_generation
= generation
;
305 fuse_init_inode(inode
, attr
);
306 unlock_new_inode(inode
);
307 } else if ((inode
->i_mode
^ attr
->mode
) & S_IFMT
) {
308 /* Inode has changed type, any I/O on the old should fail */
309 fuse_make_bad(inode
);
314 fi
= get_fuse_inode(inode
);
315 spin_lock(&fi
->lock
);
317 spin_unlock(&fi
->lock
);
318 fuse_change_attributes(inode
, attr
, attr_valid
, attr_version
);
323 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
324 loff_t offset
, loff_t len
)
326 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
327 struct fuse_inode
*fi
;
332 inode
= ilookup5(sb
, nodeid
, fuse_inode_eq
, &nodeid
);
336 fi
= get_fuse_inode(inode
);
337 spin_lock(&fi
->lock
);
338 fi
->attr_version
= atomic64_inc_return(&fc
->attr_version
);
339 spin_unlock(&fi
->lock
);
341 fuse_invalidate_attr(inode
);
342 forget_all_cached_acls(inode
);
344 pg_start
= offset
>> PAGE_SHIFT
;
348 pg_end
= (offset
+ len
- 1) >> PAGE_SHIFT
;
349 invalidate_inode_pages2_range(inode
->i_mapping
,
356 bool fuse_lock_inode(struct inode
*inode
)
360 if (!get_fuse_conn(inode
)->parallel_dirops
) {
361 mutex_lock(&get_fuse_inode(inode
)->mutex
);
368 void fuse_unlock_inode(struct inode
*inode
, bool locked
)
371 mutex_unlock(&get_fuse_inode(inode
)->mutex
);
374 static void fuse_umount_begin(struct super_block
*sb
)
376 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
378 if (!fc
->no_force_umount
)
382 static void fuse_send_destroy(struct fuse_conn
*fc
)
387 args
.opcode
= FUSE_DESTROY
;
390 fuse_simple_request(fc
, &args
);
394 static void fuse_put_super(struct super_block
*sb
)
396 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
398 mutex_lock(&fuse_mutex
);
399 list_del(&fc
->entry
);
400 fuse_ctl_remove_conn(fc
);
401 mutex_unlock(&fuse_mutex
);
406 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
408 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
409 stbuf
->f_bsize
= attr
->bsize
;
410 stbuf
->f_frsize
= attr
->frsize
;
411 stbuf
->f_blocks
= attr
->blocks
;
412 stbuf
->f_bfree
= attr
->bfree
;
413 stbuf
->f_bavail
= attr
->bavail
;
414 stbuf
->f_files
= attr
->files
;
415 stbuf
->f_ffree
= attr
->ffree
;
416 stbuf
->f_namelen
= attr
->namelen
;
417 /* fsid is left zero */
420 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
422 struct super_block
*sb
= dentry
->d_sb
;
423 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
425 struct fuse_statfs_out outarg
;
428 if (!fuse_allow_current_process(fc
)) {
429 buf
->f_type
= FUSE_SUPER_MAGIC
;
433 memset(&outarg
, 0, sizeof(outarg
));
435 args
.opcode
= FUSE_STATFS
;
436 args
.nodeid
= get_node_id(d_inode(dentry
));
437 args
.out_numargs
= 1;
438 args
.out_args
[0].size
= sizeof(outarg
);
439 args
.out_args
[0].value
= &outarg
;
440 err
= fuse_simple_request(fc
, &args
);
442 convert_fuse_statfs(buf
, &outarg
.st
);
453 OPT_DEFAULT_PERMISSIONS
,
460 static const struct fs_parameter_spec fuse_param_specs
[] = {
461 fsparam_string ("source", OPT_SOURCE
),
462 fsparam_u32 ("fd", OPT_FD
),
463 fsparam_u32oct ("rootmode", OPT_ROOTMODE
),
464 fsparam_u32 ("user_id", OPT_USER_ID
),
465 fsparam_u32 ("group_id", OPT_GROUP_ID
),
466 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS
),
467 fsparam_flag ("allow_other", OPT_ALLOW_OTHER
),
468 fsparam_u32 ("max_read", OPT_MAX_READ
),
469 fsparam_u32 ("blksize", OPT_BLKSIZE
),
470 fsparam_string ("subtype", OPT_SUBTYPE
),
474 static const struct fs_parameter_description fuse_fs_parameters
= {
476 .specs
= fuse_param_specs
,
479 static int fuse_parse_param(struct fs_context
*fc
, struct fs_parameter
*param
)
481 struct fs_parse_result result
;
482 struct fuse_fs_context
*ctx
= fc
->fs_private
;
486 * Ignore options coming from mount(MS_REMOUNT) for backward
489 if (fc
->purpose
== FS_CONTEXT_FOR_RECONFIGURE
)
492 opt
= fs_parse(fc
, &fuse_fs_parameters
, param
, &result
);
499 return invalf(fc
, "fuse: Multiple sources specified");
500 fc
->source
= param
->string
;
501 param
->string
= NULL
;
506 return invalf(fc
, "fuse: Multiple subtypes specified");
507 ctx
->subtype
= param
->string
;
508 param
->string
= NULL
;
512 ctx
->fd
= result
.uint_32
;
517 if (!fuse_valid_type(result
.uint_32
))
518 return invalf(fc
, "fuse: Invalid rootmode");
519 ctx
->rootmode
= result
.uint_32
;
520 ctx
->rootmode_present
= 1;
524 ctx
->user_id
= make_kuid(fc
->user_ns
, result
.uint_32
);
525 if (!uid_valid(ctx
->user_id
))
526 return invalf(fc
, "fuse: Invalid user_id");
527 ctx
->user_id_present
= 1;
531 ctx
->group_id
= make_kgid(fc
->user_ns
, result
.uint_32
);
532 if (!gid_valid(ctx
->group_id
))
533 return invalf(fc
, "fuse: Invalid group_id");
534 ctx
->group_id_present
= 1;
537 case OPT_DEFAULT_PERMISSIONS
:
538 ctx
->default_permissions
= 1;
541 case OPT_ALLOW_OTHER
:
542 ctx
->allow_other
= 1;
546 ctx
->max_read
= result
.uint_32
;
551 return invalf(fc
, "fuse: blksize only supported for fuseblk");
552 ctx
->blksize
= result
.uint_32
;
562 static void fuse_free_fc(struct fs_context
*fc
)
564 struct fuse_fs_context
*ctx
= fc
->fs_private
;
572 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
574 struct super_block
*sb
= root
->d_sb
;
575 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
577 if (fc
->no_mount_options
)
580 seq_printf(m
, ",user_id=%u", from_kuid_munged(fc
->user_ns
, fc
->user_id
));
581 seq_printf(m
, ",group_id=%u", from_kgid_munged(fc
->user_ns
, fc
->group_id
));
582 if (fc
->default_permissions
)
583 seq_puts(m
, ",default_permissions");
585 seq_puts(m
, ",allow_other");
586 if (fc
->max_read
!= ~0)
587 seq_printf(m
, ",max_read=%u", fc
->max_read
);
588 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
589 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
593 static void fuse_iqueue_init(struct fuse_iqueue
*fiq
,
594 const struct fuse_iqueue_ops
*ops
,
597 memset(fiq
, 0, sizeof(struct fuse_iqueue
));
598 spin_lock_init(&fiq
->lock
);
599 init_waitqueue_head(&fiq
->waitq
);
600 INIT_LIST_HEAD(&fiq
->pending
);
601 INIT_LIST_HEAD(&fiq
->interrupts
);
602 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
608 static void fuse_pqueue_init(struct fuse_pqueue
*fpq
)
612 spin_lock_init(&fpq
->lock
);
613 for (i
= 0; i
< FUSE_PQ_HASH_SIZE
; i
++)
614 INIT_LIST_HEAD(&fpq
->processing
[i
]);
615 INIT_LIST_HEAD(&fpq
->io
);
619 void fuse_conn_init(struct fuse_conn
*fc
, struct user_namespace
*user_ns
,
620 const struct fuse_iqueue_ops
*fiq_ops
, void *fiq_priv
)
622 memset(fc
, 0, sizeof(*fc
));
623 spin_lock_init(&fc
->lock
);
624 spin_lock_init(&fc
->bg_lock
);
625 init_rwsem(&fc
->killsb
);
626 refcount_set(&fc
->count
, 1);
627 atomic_set(&fc
->dev_count
, 1);
628 init_waitqueue_head(&fc
->blocked_waitq
);
629 fuse_iqueue_init(&fc
->iq
, fiq_ops
, fiq_priv
);
630 INIT_LIST_HEAD(&fc
->bg_queue
);
631 INIT_LIST_HEAD(&fc
->entry
);
632 INIT_LIST_HEAD(&fc
->devices
);
633 atomic_set(&fc
->num_waiting
, 0);
634 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
635 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
636 atomic64_set(&fc
->khctr
, 0);
637 fc
->polled_files
= RB_ROOT
;
641 atomic64_set(&fc
->attr_version
, 1);
642 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
643 fc
->pid_ns
= get_pid_ns(task_active_pid_ns(current
));
644 fc
->user_ns
= get_user_ns(user_ns
);
645 fc
->max_pages
= FUSE_DEFAULT_MAX_PAGES_PER_REQ
;
647 EXPORT_SYMBOL_GPL(fuse_conn_init
);
649 void fuse_conn_put(struct fuse_conn
*fc
)
651 if (refcount_dec_and_test(&fc
->count
)) {
652 struct fuse_iqueue
*fiq
= &fc
->iq
;
654 if (fiq
->ops
->release
)
655 fiq
->ops
->release(fiq
);
656 put_pid_ns(fc
->pid_ns
);
657 put_user_ns(fc
->user_ns
);
661 EXPORT_SYMBOL_GPL(fuse_conn_put
);
663 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
665 refcount_inc(&fc
->count
);
668 EXPORT_SYMBOL_GPL(fuse_conn_get
);
670 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
672 struct fuse_attr attr
;
673 memset(&attr
, 0, sizeof(attr
));
676 attr
.ino
= FUSE_ROOT_ID
;
678 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
681 struct fuse_inode_handle
{
686 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
687 struct fuse_inode_handle
*handle
)
689 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
691 struct dentry
*entry
;
694 if (handle
->nodeid
== 0)
697 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
699 struct fuse_entry_out outarg
;
700 const struct qstr name
= QSTR_INIT(".", 1);
702 if (!fc
->export_support
)
705 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
707 if (err
&& err
!= -ENOENT
)
714 if (get_node_id(inode
) != handle
->nodeid
)
718 if (inode
->i_generation
!= handle
->generation
)
721 entry
= d_obtain_alias(inode
);
722 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
723 fuse_invalidate_entry_cache(entry
);
733 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
734 struct inode
*parent
)
736 int len
= parent
? 6 : 3;
740 if (*max_len
< len
) {
742 return FILEID_INVALID
;
745 nodeid
= get_fuse_inode(inode
)->nodeid
;
746 generation
= inode
->i_generation
;
748 fh
[0] = (u32
)(nodeid
>> 32);
749 fh
[1] = (u32
)(nodeid
& 0xffffffff);
753 nodeid
= get_fuse_inode(parent
)->nodeid
;
754 generation
= parent
->i_generation
;
756 fh
[3] = (u32
)(nodeid
>> 32);
757 fh
[4] = (u32
)(nodeid
& 0xffffffff);
762 return parent
? 0x82 : 0x81;
765 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
766 struct fid
*fid
, int fh_len
, int fh_type
)
768 struct fuse_inode_handle handle
;
770 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
773 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
774 handle
.nodeid
|= (u64
) fid
->raw
[1];
775 handle
.generation
= fid
->raw
[2];
776 return fuse_get_dentry(sb
, &handle
);
779 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
780 struct fid
*fid
, int fh_len
, int fh_type
)
782 struct fuse_inode_handle parent
;
784 if (fh_type
!= 0x82 || fh_len
< 6)
787 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
788 parent
.nodeid
|= (u64
) fid
->raw
[4];
789 parent
.generation
= fid
->raw
[5];
790 return fuse_get_dentry(sb
, &parent
);
793 static struct dentry
*fuse_get_parent(struct dentry
*child
)
795 struct inode
*child_inode
= d_inode(child
);
796 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
798 struct dentry
*parent
;
799 struct fuse_entry_out outarg
;
800 const struct qstr name
= QSTR_INIT("..", 2);
803 if (!fc
->export_support
)
804 return ERR_PTR(-ESTALE
);
806 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
807 &name
, &outarg
, &inode
);
810 return ERR_PTR(-ESTALE
);
814 parent
= d_obtain_alias(inode
);
815 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
816 fuse_invalidate_entry_cache(parent
);
821 static const struct export_operations fuse_export_operations
= {
822 .fh_to_dentry
= fuse_fh_to_dentry
,
823 .fh_to_parent
= fuse_fh_to_parent
,
824 .encode_fh
= fuse_encode_fh
,
825 .get_parent
= fuse_get_parent
,
828 static const struct super_operations fuse_super_operations
= {
829 .alloc_inode
= fuse_alloc_inode
,
830 .free_inode
= fuse_free_inode
,
831 .evict_inode
= fuse_evict_inode
,
832 .write_inode
= fuse_write_inode
,
833 .drop_inode
= generic_delete_inode
,
834 .put_super
= fuse_put_super
,
835 .umount_begin
= fuse_umount_begin
,
836 .statfs
= fuse_statfs
,
837 .show_options
= fuse_show_options
,
840 static void sanitize_global_limit(unsigned *limit
)
843 * The default maximum number of async requests is calculated to consume
844 * 1/2^13 of the total memory, assuming 392 bytes per request.
847 *limit
= ((totalram_pages() << PAGE_SHIFT
) >> 13) / 392;
849 if (*limit
>= 1 << 16)
850 *limit
= (1 << 16) - 1;
853 static int set_global_limit(const char *val
, const struct kernel_param
*kp
)
857 rv
= param_set_uint(val
, kp
);
861 sanitize_global_limit((unsigned *)kp
->arg
);
866 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
868 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
873 sanitize_global_limit(&max_user_bgreq
);
874 sanitize_global_limit(&max_user_congthresh
);
876 spin_lock(&fc
->bg_lock
);
877 if (arg
->max_background
) {
878 fc
->max_background
= arg
->max_background
;
880 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
881 fc
->max_background
= max_user_bgreq
;
883 if (arg
->congestion_threshold
) {
884 fc
->congestion_threshold
= arg
->congestion_threshold
;
886 if (!cap_sys_admin
&&
887 fc
->congestion_threshold
> max_user_congthresh
)
888 fc
->congestion_threshold
= max_user_congthresh
;
890 spin_unlock(&fc
->bg_lock
);
893 struct fuse_init_args
{
894 struct fuse_args args
;
895 struct fuse_init_in in
;
896 struct fuse_init_out out
;
899 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_args
*args
,
902 struct fuse_init_args
*ia
= container_of(args
, typeof(*ia
), args
);
903 struct fuse_init_out
*arg
= &ia
->out
;
905 if (error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
908 unsigned long ra_pages
;
910 process_init_limits(fc
, arg
);
912 if (arg
->minor
>= 6) {
913 ra_pages
= arg
->max_readahead
/ PAGE_SIZE
;
914 if (arg
->flags
& FUSE_ASYNC_READ
)
916 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
918 if (arg
->minor
>= 17) {
919 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
922 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
925 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
926 fc
->atomic_o_trunc
= 1;
927 if (arg
->minor
>= 9) {
928 /* LOOKUP has dependency on proto version */
929 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
930 fc
->export_support
= 1;
932 if (arg
->flags
& FUSE_BIG_WRITES
)
934 if (arg
->flags
& FUSE_DONT_MASK
)
936 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
937 fc
->auto_inval_data
= 1;
938 else if (arg
->flags
& FUSE_EXPLICIT_INVAL_DATA
)
939 fc
->explicit_inval_data
= 1;
940 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
941 fc
->do_readdirplus
= 1;
942 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
943 fc
->readdirplus_auto
= 1;
945 if (arg
->flags
& FUSE_ASYNC_DIO
)
947 if (arg
->flags
& FUSE_WRITEBACK_CACHE
)
948 fc
->writeback_cache
= 1;
949 if (arg
->flags
& FUSE_PARALLEL_DIROPS
)
950 fc
->parallel_dirops
= 1;
951 if (arg
->flags
& FUSE_HANDLE_KILLPRIV
)
952 fc
->handle_killpriv
= 1;
953 if (arg
->time_gran
&& arg
->time_gran
<= 1000000000)
954 fc
->sb
->s_time_gran
= arg
->time_gran
;
955 if ((arg
->flags
& FUSE_POSIX_ACL
)) {
956 fc
->default_permissions
= 1;
958 fc
->sb
->s_xattr
= fuse_acl_xattr_handlers
;
960 if (arg
->flags
& FUSE_CACHE_SYMLINKS
)
961 fc
->cache_symlinks
= 1;
962 if (arg
->flags
& FUSE_ABORT_ERROR
)
964 if (arg
->flags
& FUSE_MAX_PAGES
) {
966 min_t(unsigned int, FUSE_MAX_MAX_PAGES
,
967 max_t(unsigned int, arg
->max_pages
, 1));
970 ra_pages
= fc
->max_read
/ PAGE_SIZE
;
975 fc
->sb
->s_bdi
->ra_pages
=
976 min(fc
->sb
->s_bdi
->ra_pages
, ra_pages
);
977 fc
->minor
= arg
->minor
;
978 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
979 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
984 fuse_set_initialized(fc
);
985 wake_up_all(&fc
->blocked_waitq
);
988 void fuse_send_init(struct fuse_conn
*fc
)
990 struct fuse_init_args
*ia
;
992 ia
= kzalloc(sizeof(*ia
), GFP_KERNEL
| __GFP_NOFAIL
);
994 ia
->in
.major
= FUSE_KERNEL_VERSION
;
995 ia
->in
.minor
= FUSE_KERNEL_MINOR_VERSION
;
996 ia
->in
.max_readahead
= fc
->sb
->s_bdi
->ra_pages
* PAGE_SIZE
;
998 FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
999 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
1000 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
1001 FUSE_FLOCK_LOCKS
| FUSE_HAS_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
1002 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
|
1003 FUSE_WRITEBACK_CACHE
| FUSE_NO_OPEN_SUPPORT
|
1004 FUSE_PARALLEL_DIROPS
| FUSE_HANDLE_KILLPRIV
| FUSE_POSIX_ACL
|
1005 FUSE_ABORT_ERROR
| FUSE_MAX_PAGES
| FUSE_CACHE_SYMLINKS
|
1006 FUSE_NO_OPENDIR_SUPPORT
| FUSE_EXPLICIT_INVAL_DATA
;
1007 ia
->args
.opcode
= FUSE_INIT
;
1008 ia
->args
.in_numargs
= 1;
1009 ia
->args
.in_args
[0].size
= sizeof(ia
->in
);
1010 ia
->args
.in_args
[0].value
= &ia
->in
;
1011 ia
->args
.out_numargs
= 1;
1012 /* Variable length argument used for backward compatibility
1013 with interface version < 7.5. Rest of init_out is zeroed
1014 by do_get_request(), so a short reply is not a problem */
1015 ia
->args
.out_argvar
= 1;
1016 ia
->args
.out_args
[0].size
= sizeof(ia
->out
);
1017 ia
->args
.out_args
[0].value
= &ia
->out
;
1018 ia
->args
.force
= true;
1019 ia
->args
.nocreds
= true;
1020 ia
->args
.end
= process_init_reply
;
1022 if (fuse_simple_background(fc
, &ia
->args
, GFP_KERNEL
) != 0)
1023 process_init_reply(fc
, &ia
->args
, -ENOTCONN
);
1025 EXPORT_SYMBOL_GPL(fuse_send_init
);
1027 void fuse_free_conn(struct fuse_conn
*fc
)
1029 WARN_ON(!list_empty(&fc
->devices
));
1032 EXPORT_SYMBOL_GPL(fuse_free_conn
);
1034 static int fuse_bdi_init(struct fuse_conn
*fc
, struct super_block
*sb
)
1040 suffix
= "-fuseblk";
1042 * sb->s_bdi points to blkdev's bdi however we want to redirect
1043 * it to our private bdi...
1046 sb
->s_bdi
= &noop_backing_dev_info
;
1048 err
= super_setup_bdi_name(sb
, "%u:%u%s", MAJOR(fc
->dev
),
1049 MINOR(fc
->dev
), suffix
);
1053 sb
->s_bdi
->ra_pages
= VM_READAHEAD_PAGES
;
1054 /* fuse does it's own writeback accounting */
1055 sb
->s_bdi
->capabilities
= BDI_CAP_NO_ACCT_WB
| BDI_CAP_STRICTLIMIT
;
1058 * For a single fuse filesystem use max 1% of dirty +
1059 * writeback threshold.
1061 * This gives about 1M of write buffer for memory maps on a
1062 * machine with 1G and 10% dirty_ratio, which should be more
1065 * Privileged users can raise it by writing to
1067 * /sys/class/bdi/<bdi>/max_ratio
1069 bdi_set_max_ratio(sb
->s_bdi
, 1);
1074 struct fuse_dev
*fuse_dev_alloc(void)
1076 struct fuse_dev
*fud
;
1077 struct list_head
*pq
;
1079 fud
= kzalloc(sizeof(struct fuse_dev
), GFP_KERNEL
);
1083 pq
= kcalloc(FUSE_PQ_HASH_SIZE
, sizeof(struct list_head
), GFP_KERNEL
);
1089 fud
->pq
.processing
= pq
;
1090 fuse_pqueue_init(&fud
->pq
);
1094 EXPORT_SYMBOL_GPL(fuse_dev_alloc
);
1096 void fuse_dev_install(struct fuse_dev
*fud
, struct fuse_conn
*fc
)
1098 fud
->fc
= fuse_conn_get(fc
);
1099 spin_lock(&fc
->lock
);
1100 list_add_tail(&fud
->entry
, &fc
->devices
);
1101 spin_unlock(&fc
->lock
);
1103 EXPORT_SYMBOL_GPL(fuse_dev_install
);
1105 struct fuse_dev
*fuse_dev_alloc_install(struct fuse_conn
*fc
)
1107 struct fuse_dev
*fud
;
1109 fud
= fuse_dev_alloc();
1113 fuse_dev_install(fud
, fc
);
1116 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install
);
1118 void fuse_dev_free(struct fuse_dev
*fud
)
1120 struct fuse_conn
*fc
= fud
->fc
;
1123 spin_lock(&fc
->lock
);
1124 list_del(&fud
->entry
);
1125 spin_unlock(&fc
->lock
);
1129 kfree(fud
->pq
.processing
);
1132 EXPORT_SYMBOL_GPL(fuse_dev_free
);
1134 int fuse_fill_super_common(struct super_block
*sb
, struct fuse_fs_context
*ctx
)
1136 struct fuse_dev
*fud
;
1137 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1139 struct dentry
*root_dentry
;
1143 if (sb
->s_flags
& SB_MANDLOCK
)
1146 sb
->s_flags
&= ~(SB_NOSEC
| SB_I_VERSION
);
1151 if (!sb_set_blocksize(sb
, ctx
->blksize
))
1155 sb
->s_blocksize
= PAGE_SIZE
;
1156 sb
->s_blocksize_bits
= PAGE_SHIFT
;
1159 sb
->s_subtype
= ctx
->subtype
;
1160 ctx
->subtype
= NULL
;
1161 sb
->s_magic
= FUSE_SUPER_MAGIC
;
1162 sb
->s_op
= &fuse_super_operations
;
1163 sb
->s_xattr
= fuse_xattr_handlers
;
1164 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1165 sb
->s_time_gran
= 1;
1166 sb
->s_export_op
= &fuse_export_operations
;
1167 sb
->s_iflags
|= SB_I_IMA_UNVERIFIABLE_SIGNATURE
;
1168 if (sb
->s_user_ns
!= &init_user_ns
)
1169 sb
->s_iflags
|= SB_I_UNTRUSTED_MOUNTER
;
1172 * If we are not in the initial user namespace posix
1173 * acls must be translated.
1175 if (sb
->s_user_ns
!= &init_user_ns
)
1176 sb
->s_xattr
= fuse_no_acl_xattr_handlers
;
1178 fud
= fuse_dev_alloc_install(fc
);
1182 fc
->dev
= sb
->s_dev
;
1184 err
= fuse_bdi_init(fc
, sb
);
1188 /* Handle umasking inside the fuse code */
1189 if (sb
->s_flags
& SB_POSIXACL
)
1191 sb
->s_flags
|= SB_POSIXACL
;
1193 fc
->default_permissions
= ctx
->default_permissions
;
1194 fc
->allow_other
= ctx
->allow_other
;
1195 fc
->user_id
= ctx
->user_id
;
1196 fc
->group_id
= ctx
->group_id
;
1197 fc
->max_read
= max_t(unsigned, 4096, ctx
->max_read
);
1198 fc
->destroy
= ctx
->destroy
;
1199 fc
->no_control
= ctx
->no_control
;
1200 fc
->no_force_umount
= ctx
->no_force_umount
;
1201 fc
->no_mount_options
= ctx
->no_mount_options
;
1204 root
= fuse_get_root_inode(sb
, ctx
->rootmode
);
1205 sb
->s_d_op
= &fuse_root_dentry_operations
;
1206 root_dentry
= d_make_root(root
);
1209 /* Root dentry doesn't have .d_revalidate */
1210 sb
->s_d_op
= &fuse_dentry_operations
;
1212 mutex_lock(&fuse_mutex
);
1217 err
= fuse_ctl_add_conn(fc
);
1221 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1222 sb
->s_root
= root_dentry
;
1224 mutex_unlock(&fuse_mutex
);
1228 mutex_unlock(&fuse_mutex
);
1235 EXPORT_SYMBOL_GPL(fuse_fill_super_common
);
1237 static int fuse_fill_super(struct super_block
*sb
, struct fs_context
*fsc
)
1239 struct fuse_fs_context
*ctx
= fsc
->fs_private
;
1242 struct fuse_conn
*fc
;
1245 file
= fget(ctx
->fd
);
1250 * Require mount to happen from the same user namespace which
1251 * opened /dev/fuse to prevent potential attacks.
1253 if ((file
->f_op
!= &fuse_dev_operations
) ||
1254 (file
->f_cred
->user_ns
!= sb
->s_user_ns
))
1256 ctx
->fudptr
= &file
->private_data
;
1258 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1263 fuse_conn_init(fc
, sb
->s_user_ns
, &fuse_dev_fiq_ops
, NULL
);
1264 fc
->release
= fuse_free_conn
;
1267 err
= fuse_fill_super_common(sb
, ctx
);
1271 * atomic_dec_and_test() in fput() provides the necessary
1272 * memory barrier for file->private_data to be visible on all
1276 fuse_send_init(get_fuse_conn_super(sb
));
1281 sb
->s_fs_info
= NULL
;
1288 static int fuse_get_tree(struct fs_context
*fc
)
1290 struct fuse_fs_context
*ctx
= fc
->fs_private
;
1292 if (!ctx
->fd_present
|| !ctx
->rootmode_present
||
1293 !ctx
->user_id_present
|| !ctx
->group_id_present
)
1298 return get_tree_bdev(fc
, fuse_fill_super
);
1301 return get_tree_nodev(fc
, fuse_fill_super
);
1304 static const struct fs_context_operations fuse_context_ops
= {
1305 .free
= fuse_free_fc
,
1306 .parse_param
= fuse_parse_param
,
1307 .reconfigure
= fuse_reconfigure
,
1308 .get_tree
= fuse_get_tree
,
1312 * Set up the filesystem mount context.
1314 static int fuse_init_fs_context(struct fs_context
*fc
)
1316 struct fuse_fs_context
*ctx
;
1318 ctx
= kzalloc(sizeof(struct fuse_fs_context
), GFP_KERNEL
);
1323 ctx
->blksize
= FUSE_DEFAULT_BLKSIZE
;
1326 if (fc
->fs_type
== &fuseblk_fs_type
) {
1327 ctx
->is_bdev
= true;
1328 ctx
->destroy
= true;
1332 fc
->fs_private
= ctx
;
1333 fc
->ops
= &fuse_context_ops
;
1337 static void fuse_sb_destroy(struct super_block
*sb
)
1339 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1343 fuse_send_destroy(fc
);
1345 fuse_abort_conn(fc
);
1346 fuse_wait_aborted(fc
);
1348 down_write(&fc
->killsb
);
1350 up_write(&fc
->killsb
);
1354 void fuse_kill_sb_anon(struct super_block
*sb
)
1356 fuse_sb_destroy(sb
);
1357 kill_anon_super(sb
);
1359 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon
);
1361 static struct file_system_type fuse_fs_type
= {
1362 .owner
= THIS_MODULE
,
1364 .fs_flags
= FS_HAS_SUBTYPE
| FS_USERNS_MOUNT
,
1365 .init_fs_context
= fuse_init_fs_context
,
1366 .parameters
= &fuse_fs_parameters
,
1367 .kill_sb
= fuse_kill_sb_anon
,
1369 MODULE_ALIAS_FS("fuse");
1372 static void fuse_kill_sb_blk(struct super_block
*sb
)
1374 fuse_sb_destroy(sb
);
1375 kill_block_super(sb
);
1378 static struct file_system_type fuseblk_fs_type
= {
1379 .owner
= THIS_MODULE
,
1381 .init_fs_context
= fuse_init_fs_context
,
1382 .parameters
= &fuse_fs_parameters
,
1383 .kill_sb
= fuse_kill_sb_blk
,
1384 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1386 MODULE_ALIAS_FS("fuseblk");
1388 static inline int register_fuseblk(void)
1390 return register_filesystem(&fuseblk_fs_type
);
1393 static inline void unregister_fuseblk(void)
1395 unregister_filesystem(&fuseblk_fs_type
);
1398 static inline int register_fuseblk(void)
1403 static inline void unregister_fuseblk(void)
1408 static void fuse_inode_init_once(void *foo
)
1410 struct inode
*inode
= foo
;
1412 inode_init_once(inode
);
1415 static int __init
fuse_fs_init(void)
1419 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1420 sizeof(struct fuse_inode
), 0,
1421 SLAB_HWCACHE_ALIGN
|SLAB_ACCOUNT
|SLAB_RECLAIM_ACCOUNT
,
1422 fuse_inode_init_once
);
1424 if (!fuse_inode_cachep
)
1427 err
= register_fuseblk();
1431 err
= register_filesystem(&fuse_fs_type
);
1438 unregister_fuseblk();
1440 kmem_cache_destroy(fuse_inode_cachep
);
1445 static void fuse_fs_cleanup(void)
1447 unregister_filesystem(&fuse_fs_type
);
1448 unregister_fuseblk();
1451 * Make sure all delayed rcu free inodes are flushed before we
1455 kmem_cache_destroy(fuse_inode_cachep
);
1458 static struct kobject
*fuse_kobj
;
1460 static int fuse_sysfs_init(void)
1464 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1470 err
= sysfs_create_mount_point(fuse_kobj
, "connections");
1472 goto out_fuse_unregister
;
1476 out_fuse_unregister
:
1477 kobject_put(fuse_kobj
);
1482 static void fuse_sysfs_cleanup(void)
1484 sysfs_remove_mount_point(fuse_kobj
, "connections");
1485 kobject_put(fuse_kobj
);
1488 static int __init
fuse_init(void)
1492 pr_info("init (API version %i.%i)\n",
1493 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1495 INIT_LIST_HEAD(&fuse_conn_list
);
1496 res
= fuse_fs_init();
1500 res
= fuse_dev_init();
1502 goto err_fs_cleanup
;
1504 res
= fuse_sysfs_init();
1506 goto err_dev_cleanup
;
1508 res
= fuse_ctl_init();
1510 goto err_sysfs_cleanup
;
1512 sanitize_global_limit(&max_user_bgreq
);
1513 sanitize_global_limit(&max_user_congthresh
);
1518 fuse_sysfs_cleanup();
1527 static void __exit
fuse_exit(void)
1532 fuse_sysfs_cleanup();
1537 module_init(fuse_init
);
1538 module_exit(fuse_exit
);